AI agents and automation
Custom AI agents for your business.
Built around your business, not around the rules of a platform.
We advise mid-sized companies and online retailers on AI agents AI agent An AI agent is a software system, built on a large language model, that carries out actions on its own and uses interfaces to other programs to reach its goal. Read in the glossary → and automations, then build and run them around your processes, systems and data sources. The system, the workflows and the data run on your infrastructure and belong to you. The models themselves run externally, and we choose them to suit the task.
Why bespoke
A market that is changing fast.
AI agents are a young discipline. Take the Model Context Protocol MCP MCP, the Model Context Protocol, is an open protocol that connects applications built on large language models with external data sources and tools. Read in the glossary → (MCP), the open protocol that agents use to connect to tools and data: the current version of its specification is dated 28 July 2026. Meanwhile, one platform vendor is in the middle of folding its agent product into another of its offerings.
Some limits the platforms point out themselves:
- In the cloud version of n8n n8n n8n is a tool for automating business workflows, including steps that use AI; it is published under fair-code licences and can run on a company's own infrastructure. Read in the glossary → , the Code node can't import external npm modules.
- n8n can make workflows available as an MCP server, but currently not over stdio (standard input and output).
- Another vendor advises against using its agents with sensitive data, for high-stakes financial decisions or wherever strict legal requirements apply.
Experience has taught us that our own pipelines run better without n8n. What we build with skills, subagents and a deliberate choice of model couldn't be replicated in n8n.
We still use n8n where it fits. One example from pharmaceutical logistics Pharma logistics Pharma logistics is the term we use for storing, transporting and delivering medicinal products and other pharma-sector goods, such as promotional material and supplies for patients. Read in the glossary → : in a business application Business application In general terms, a business application is software used to perform or support business functions. Read in the glossary → we built, we took the AI AI Artificial intelligence (AI) covers methods that let a computer perform tasks which would require intelligence if a person did them. Read in the glossary → functions off n8n in August 2026. They now call the model directly through its provider's SDK. Only the chatbot stayed on n8n, and that was a deliberate choice. Today we combine whatever the job needs: the model providers' SDKs and their agent tooling, and MCP. For special cases, locally hosted models or cloud agents are an option too.
How we build agents
Three principles, and a person has the final say.
Our agents produce drafts and previews; nothing is published until a person has approved it. Along the way, we stick to three rules:
Separate roles, independent checks
The agent that writes doesn't do the research, and the one that checks has done neither. In one of our pipelines, for example, the checker runs on a different model and knows nothing of the steps before it. On top of that, a script with no AI involved looks for every quote, word for word, in its source.
The right model for each step
In one system we built, a small model handles the routine work, a mid-sized one the long prompts and the largest the demanding analysis. The assignment can be changed in the back end without taking the system offline. When a better model arrives, we switch.
A dedicated API and MCP as the interface
Agents work with the application rather than around it, for example through MCP tools and a dedicated API secured by tokens. In one CMS project, the server checks every change an agent makes against the permitted building blocks, rejects anything that doesn't match and says what needs to change.
What we build
Agents that do one job properly and leave a trail.
Six kinds of work we build agents and automations for, most of them with an example from our own projects. Our case studies from pharmaceutical logistics and e-commerce follow individual projects from start to finish.
-
01
Research and checking pipelines with a source for every claim
Here is how one of our pipelines works: an agent gathers the facts and backs every entry with a verbatim quote and its source. A second, independent agent then retrieves each source itself and checks that the passage is really there.
-
02
Email agents that guard confidentiality
The email agent in one business application, for example, sorts each incoming email and turns it into tasks and knowledge-base entries. If it finds access credentials, contracts or personal data, it raises the email's confidentiality level of its own accord and never lowers it. It can't delete anything, and the system logs every write.
-
03
Agents that plug into business applications via API and MCP
Take the REST API of one business application: its responses are compact enough to fit an agent's context window, and it supplies values the agent would otherwise have to guess, such as the next free reference number. Agents work with that interface through an MCP server, with a large model doing the analysis and a smaller one making all the calls.
-
04
Reports with figures straight from your data
In one application's monthly AI status report, for example, the model writes only the text: the summary, assessment, recommendations and closing remarks. The template takes the figures directly from the data.
-
05
Content from product data
This is part of what we do as well. How such content works with search engines and AI-generated answers is covered on our SEO and content from product data page.
-
06
ERP, shop and other systems in sync
Anything that follows fixed rules doesn't need an agent, and one platform vendor gives the same advice. How we connect your shop is explained on our online shops and shop processes page.
We've been working with total10 for many years, and it's always been fast and effective. t10 looks after and builds everything our various shops need. Lately, AI has been coming up more and more. What total10 makes possible there goes far beyond anything we'd imagined ourselves. Anyone on board now is years ahead of the competition. Anyone who isn't will be left behind.
From idea to running agent
-
Step 1
Use case and data
- what is done by hand today
- which systems
- which data
-
Step 2
Architecture and model choice
- agent roles
- a model for each step
- review stages
-
Step 3
Integration
- APIs
- MCP
- data sources
-
Step 4
Testing and approval
- test runs
- automated checks
- sign-off by a person
-
Step 5
Operation and development
- a log of every run
- model changes
- new tasks
Agents with your data
When an integration isn't enough.
Some agents need business logic, permissions and endpoints of their own. In that case we build the application to go with them, as part of our bespoke Symfony development.
Business logic stays in the application
In one project, the application assembles the prompt cascade for its chatbot itself, and n8n simply passes the finished prompt on. Its email agent runs in the application's own process, and the agent's tools work directly on the data layer.
A record of what really happened
With every reply, that email agent delivers a record of the actions it actually carried out. The record is written by the system, not by the model, so if the model reports a success that never happened, it shows. For every run, you can trace the tool calls and the token usage.
Our work
AI and automation in practice.
-
Case study · E-commerce
The self-driving content machine
An established mid-market online shop with several thousand specialist products that need real explanation, multiple country shops and a large trade magazine faced a scaling problem: high-quality, search-strong content, complete with imagery, internal linking and multiple languages, can't be produced by hand at the volume required. We built a fully automated content machine for it that generates, illustrates, translates and even voices articles, running autonomously round the clock.
Read the case study
When we advise against AI
Sometimes the honest answer is no agent at all.
Not every process needs an agent.
- A task that comes up once a quarter is better served by a checklist.
- If a process has no clear rules, a model won't bring order to it. It will only make mistakes faster.
- Some problems are solved more cheaply by a good form than by any agent.
We'll tell you that before we start. The most expensive automation is the one nobody needs.
Where AI does pay off, our guiding principle applies: people orchestrate, the model is an interchangeable tool, and the system belongs to you. Your workflows and data stay with you, whichever model happens to be doing the work.
Frequently asked questions
What people ask before starting an agent project.
- Where does our data live?
- On your infrastructure, together with the system and workflows, all of which belong to you. The models we use are hosted externally, so for every workflow we tell you exactly what each model gets to see. We have no models of our own and haven't trained any.
- Do you work with n8n or build bespoke systems?
- Both; it depends on the project. We use n8n where a workflow is enough, for example to connect a chatbot whose logic lives in the application. Where you need separate agents, review stages or a choice of model for each step, we write our own software, using the model providers' SDKs and MCP.
- Which models do you use?
- Whichever suits each step, from small models up to the largest. We don't tie ourselves to any one provider.
- How do we avoid lock-in?
- You own the system, and the models can be swapped. Where needed, we connect tools and data through MCP, an open protocol. Since December 2025 it has been part of the Agentic AI Foundation, a directed fund under the Linux Foundation, and the agent SDKs of several model providers support it.
- What does an agent cost?
- That depends on the process, so there's no fixed price. We start with a use-case analysis, and you then get a cost estimate that sets out its assumptions. If the numbers don't add up, we'll say so.
- What if an agent gets it wrong?
- We plan for that from day one. In a pipeline like the one described above, a second agent on a different model reviews the result, scripts take care of everything that can be checked mechanically, and a person signs off before anything is published. If a finding is still open after a round of corrections, a person makes that call too.
- Can this run locally?
- Yes, local models are an option. Whether one handles the task reliably enough is something we assess case by case.
A process worth automating?
Tell us briefly what the process is and which software is involved. We'll get back to you promptly with an honest view of whether a bespoke build is worthwhile and what it would look like.